ddl_strategy flags

ddl_strategy accepts flags in command line format. The flags can be vitess-specific, or, if unrecognized by Vitess, are passed on the underlying online schema change tools.

Vitess flags #

Vitess respects the following flags. They can be combined unless specifically indicated otherwise:

  • --allow-concurrent: allow a migration to run concurrently to other migrations, rather than queue sequentially. Some restrictions apply, see concurrent migrations.

  • --allow-zero-in-date: normally Vitess operates with a strict sql_mode. If you have columns such as my_datetime DATETIME DEFAULT '0000-00-00 00:00:00' and you wish to run DDL on these tables, Vitess will prevent the migration due to invalid values. Provide --allow-zero-in-date to allow either a fully zero-date or a zero-in-date inyour schema. See also NO_ZERO_IN_DATE and NO_ZERO_DATE documentation for sql_mode.

  • --analyze-table: for ALTER TABLE operation, and in online/vitess strategy, run an ANALYZE NO_WRITE_TO_BINLOG TABLE just before starting the migration process, to get better estimation of the number of rows on the migrated table.

  • --cut-over-threshold=<duration>": set an explicit threshold and timeout for a vitess ALTER TABLE cut-over phase. The default cut-over threshold, if not specified, is 10s. A vitess migration will not attempt to cut-over if the vstream, or replication lag, is more than the cut-over threshold. Also, during cut-over, table locks will timeout after the same cut-over threshold (aborting the operation). The allowed range 5s..30s. Too low and cut-over may never succeed because of the inherent async nature of vitess migrations. Too high and table locks will be placed for too long, effectively rendering the table inaccessible. Attempting to set a value outside the allowed range returns an error. A special case is when the user submits a 0 value, which subsequently presets the threshold to the default value of 10s. The user may override this value later on via ALTER VITESS_MIGRATION '...' CUTOVER_THRESHOLD '...'

  • --declarative: mark the migration as declarative. You will define a desired schema state by supplying CREATE and DROP statements, ad Vitess will infer how to achieve the desired schema. If need be, it will generate an ALTER migration to convert to the new schema. See declarative migrations.

  • --fast-range-rotation: when the migration runs on a table partitioned by RANGE, and the migration either runs a single DROP PARTITION or a single ADD PARTITION, and nothing other than that, then this flags instructs Vitess to run the ALTER TABLE statement directly against MySQL, as opposed to running an Online DDL with a shadow table. For DROP PARTITION, this flag is actually always desired, and will possibly become default/redundant in the future. If all conditions are indeed met, then the migration is not revertible.

  • --force-cut-over-after=<duration>: applicable to ALTER TABLE migrations in vitess strategy and on MySQL 8.0. The final step of the migration, the cut-over, involves acquiring locks on the migrated table. This operation can time out when the table is otherwise locked by the user or the app, in which case Vitess retries it later on, until successful. When --force-cut-over-after is specified, then counting the time since the very first cut-over attempt, and for any further cut-over attempt, Vitess will aggressively KILL queries and transactions that are using or locking the migrated table.

    For example, say --force-cut-over-after=2h, and that the migration takes 7h to run. Say there is constant workload/locking that prevents the migration from cutting over. The first cut-over attempt takes place at the end of the 7h run, and fails due to lock wait timeout. During the next 2 hours there will be multiuple additional attempt to cut-over, and say they all continue to fail. At the 2h mark (9h since starting the migration), give or take, Vitess runs a cut-over that KILLs existing queries and transactions on the table. This is likely to make the cut-over successful. Should this still fail, Vitess will continue to forcefully KILL queries and transactions in all additional cut-over attempts.

    See also forcing a migration cut-over

  • --in-order-completion: a migration that runs with this DDL strategy flag may only complete if no prior migrations are still pending (pending means either queued, ready or running states). --in-order-completion considers the order by which migrations were submitted. Note that --in-order-completion still allows concurrency. In fact, it is designed to work with concurrent migrations. The idea is that while many migrations may run concurrently, they must complete in-order.

    • This lets the user submit multiple migrations which may have some dependencies (for example, introduce two views, one of which reads from the other). As long as the migrations are submitted in a valid order, the user can then expect vitess to complete the migrations successfully (and in that order).
    • This strategy flag applies to any CREATE|DROP TABLE|VIEW statements, and to ALTER TABLE with vitess|online strategy.
    • It does not apply to ALTER TABLE when using the gh-ost, pt-osc, mysql, or direct strategies.
  • --postpone-completion: initiate a migration that will only cut-over per user command, i.e. will not auto-complete. This gives the user control over the time when the schema change takes effect. See postponed migrations.

    --declarative migrations are only evaluated when scheduled to run. If a migrations is both --declarative and --postpone-completion then it will remain in queued state until the user issues a ALTER VITESS_MIGRATION ... COMPLETE. If it turns out that Vitess should run the migration as an ALTER then it is only at that time that the migration starts.

  • --postpone-launch: initiate a migration that remains queued and only launches per user command. See postponed migrations.

  • --prefer-instant-ddl: where possible, apply ALGORITHM=INSTANT to the migration. This is applicable to ALTER TABLE migrations with vitess strategy. Vitess pre-computes whether the migration is eligible for INSTANT DDL. The MySQL documentation references an incomplete list of eligible changes. If applicable, vitess does not create a shadow table and the migration is not revertible.

  • --retain-artifacts=<duration>: set an explicit artifact retention for this migration. If nonzero, this value overrides the vttablet --retain_online_ddl_tables value.

  • --singleton: only allow a single pending migration to be submitted at a time. From the moment the migration is queued, and until either completion, failure or cancellation, no other new --singleton migration can be submitted. New requests will be rejected with error. --singleton works as a an exclusive lock for pending migrations. Note that this only affects migrations with --singleton flag. Migrations running without that flag are unaffected and unblocked.

  • --singleton-context: only allow migrations submitted under same context to be pending at any given time. Migrations submitted with a different context are rejected for as long as at least one of the initially submitted migrations is pending.

    It does not make sense to combine --singleton and --singleton-context.

  • --singleton-table: reject a new submission for a table or view which already has a pending migration.

Pass-through flags #

Flags unrecognized by Vitess are passed on to the underlying schema change tools. For example, a gh-ost migration can run with:

set @@ddl_strategy='gh-ost --max-load Threads_running=200'

Since Vitess knows nothing about --max-load it will pass it on as a command line argument to gh-ost. Consult gh-ost documentation for supported command line flags.

Similarly, a pt-online-schema-change migration can run with:

set @@ddl_strategy='pt-osc --null-to-not-null'

Consult pt-online-schema-change documentation for supported command line flags.

The vitess strategy (formerly known as online) uses Vitess internal mechanisms and is not a standalone command line tool. therefore, it has no particular command line flags. For internal testing/CI purposes, the vitess strategy supports --vreplication-test-suite, and this flag must not be used in production as it can have destructive consequences.